For 5G system performance, obtaining and maintaining the right balance of requirements for high power, power-added efficiency, and signal fidelity is critical. CCDF and PAPR measurements provide insights to help power amplifier designers achieve that goal.
While all parts of the 5G RF signal chain contribute to overall system performance, the transmitter power amplifier (PA) has characteristics that require careful attention. Nonlinear PA performance can be a critical factor that negatively impacts error vector magnitude (EVM) and bit-error-rate (BER).
Insufficient input back-off (IBO) causes compression leading to EVM degradation, which reduces the peak-to-average power ratio (PAPR) of OFDM/m-QAM signals. Increasing IBO will restore PAPR to the required level but with a penalty for amplifier efficiency and the costly need to use a higher-power PA. For these reasons, you need to find the optimum IBO setting point.
5G signal chain
The 5G transmitter signal chain starts with a digital baseband and beamforming processing and extends to the antenna array. Figure 1 shows the PA as the active component at the end of the line. PAs often use Doherty amplifiers to maximize efficiency.
Figure 1. This simplified block diagram of a 5G massive MIMO transmitter chain highlights power amplifiers.
PAs often use GaN technology, although other technologies, such as CMOS-SOI, are under consideration for the FR3 band. The telecom industry has its sights set on FR3, which runs roughly from 7 GHz to 24 GHz. Whatever technology is used, obtaining a balance between high output power, power-added efficiency (PAE), and signal fidelity is always a major consideration. To see how these factors interrelate, we’ll start with a look at the nature of 5G signals.
5G RF signals
5G uses orthogonal frequency division multiplexing (OFDM), where the multiple subcarriers are modulated by quadrature amplitude modulation (m-QAM) of up to 1024-QAM. Orthogonality occurs through spacing the subcarriers by the inverse of the symbol time (T). As shown in Figure 2, this ensures that subcarrier peaks align with the nulls of the other subcarriers, which prevents inter-subcarrier interference.
Figure 2. Orthogonality in OFDM signals occurs through proper subcarrier spacing.
5G’s numerology defines a range of subcarrier spacings. In the sub-6 GHz FR1 band, these are 15 kHz, 30 kHz, and 60 kHz. 30 kHz corresponds with an OFDM symbol time of 33.3 µs.
Because the subcarriers are transmitted simultaneously with a continually varying phase relationship determined by their frequency spacing, the subcarriers can sum, causing high power level peaks, as shown in Figure 3. The level of these peaks relative to the average level of the signal is characterized as the signal’s PAPR.
Figure 3. A summation of the signals shown in the time domain shows a high PAPR (solid trace).
When driving a PA with a high PAPR signal, the peaks of the signal drive into the amplifier’s non-linear region, which can result in spectral regrowth, causing adjacent-channel leakage. Input back-off (IBO) can reduce the peak levels, constraining them to the linear region. Unfortunately, doing so also reduces the average power. The amplifier no longer operates at its ideal point for maximum efficiency, as shown in Figure 4.
Figure 4. An amplifier’s point of maximum efficiency occurs just before it reaches the saturation region.
To regain some efficiency, you can apply various techniques that intentionally reduce PAPR and thus reduce IBO. You must limit the degree of PAPR reduction to that consistent with meeting EVM targets. Once the PA reaches that level, be sure that PA non-linearity doesn’t further reduce PAPR. Otherwise, that could increase EVM and lead to symbol errors, as shown in Figure 5.
Figure 5. Clipping the peaks reduces the vector length, potentially causing symbol errors.
Measuring EVM requires equipment such as signal analyzers. Because we are concerned with the impact of an amplifier’s nonlinearity on reducing the signal’s PAPR, you can use a cost-effective, direct measure of that effect.
Linearity characterization
You can apply any of several methods to characterize amplifier linearity. Two common methods are to measure the 1 dB compression point (P1db) and to measure the third-order intercept point (TOI). Both methods use CW signals and average power measurements. Because these methods use signals that don’t represent the OFDM/m-QAM signals, they won’t provide sufficient information about the response to signals with high PAPR levels.
Noise power ratio (NPR) is another measurement method for assessing amplifier linearity. It is effective as an indicator of spectral regrowth caused by nonlinearity. Because NPR uses additive white-Gaussian noise (AWGN), it’s also more representative of real-world performance. It does, however, require expensive test equipment. Strickler, Correa, and Bollendorf compare this and other methods for assessing amplifier nonlinearity.
A “Real-World” view of amplifier linearity
Getting back to the end objective, how can we assess the effect of an amplifier’s linearity, or rather non-linearity, on EVM performance? Figure 6 shows experimental results for the relationship between EVM and PAPR.
Figure 6. A plot of EVM vs. PAPR shows their relationship.
In this case, the slope of the curve is approximately 3% degradation in EVM per 1 dB reduction in PAPR. The slope may differ for different modulation schemes. Establishing the slope through making PAPR and EVM measurements early in development means, however, that you can use PAPR as a simple, quick, and cost-effective predictor of amplifier performance on EVM. This avoids making repeated EVM measurements when changing IBO or amplifier design. It also means that you can use PAPR rather than EVM in production tests, which leads to a paradigm shift in amplifier manufacturing.
Measuring PAPR and CCDF
How can you execute a practical method for measuring PAPR? You can use a high-sample-rate signal analyzer. Recall that such equipment is expensive, complicated, and occupies significant bench space for making EVM measurements.
Instead, you can use a power sensor. Available from several manufacturers, power sensors commonly use a diode as the sensing element. Diode-based average power sensors can measure a signal’s average power independent of modulation type. Because these sensors have a relatively slow response time, they don’t provide the instantaneous peak-envelope measurements required to obtain PAPR results.
Peak-power sensors are fast enough to track a signal’s power envelope and provide high-sample-rate instantaneous peak power results. They have sample rates of some 100 MSamples/sec, and when measuring repetitive signals, random interleaved sampling can yield effective sample rates of 10 GSamples/sec. That enables 100 ps time resolution.
To faithfully track the power envelope fluctuations of a wideband modulated signal, the sensor needs to have a wide video bandwidth and an associated fast rise time. In the case of a 100 MHz wide 5G FR1 channel, a sensor with less than 100 MHz video bandwidth (VBW) would not provide accurate results, whereas a sensor with, say, 165 MHz VBW would. Such sensors are available from several manufacturers.
Using peak-power sensors configured as shown in Figure 7, you can the measure RF signal’s peak, average, and minimum power at the amplifier’s input and output.
Figure 7. Using two peak-power sensors, you can simultaneously characterize a 5G power amplifier.
Figures 8a and 8b show instantaneous envelope power vs. time for signals with light and heavy signal compression. Figure 8a shows the input signal (CH1) and the output signal (CH2) from an amplifier operated mainly in its linear region. The output crest factor, another way of saying its maximum PAPR, reduces by just 0.6 dB compared to the input signal. In Figure 8b, with a reduced IBO, that difference increases to 3 dB, which indicates that the amplifier is operating further into its nonlinear region and imposing a much higher degree of compression.
Figure 8. Light compression of peaks (a) shows a 0.6 dB difference in PAPR, while (b) heavier peak compression increases PAPR to 3 dB.
Measuring the crest factor alone provides no statistical context. The complementary cumulative distribution function (CCDF) provides valuable additional information.
CCDF curves in Figure 9 show the percentage of time (Y-axis) that the PAPR (X-axis) is greater than a specific value. Figures 9a and 9b show CCDF curves for the same signals shown in Figures 8a and 8b. Plot 9a shows the results when an amplifier is operated in what is essentially its linear region for all but the highest peaks. The input signal (CH1) peaks are >9.4 dB relative to the average signal level for 0.01% of the time. The output of the amplifier (CH2) has peaks >9.2 dB relative to the average signal level for 0.01% of the time.
Figure 9. CCDF curves show that hanging the IBO, as done in Figure 8, reduces PAPR 99.99% of the time.
When the IBO has been reduced, as shown in Figure 9b, the output CCDF (CH2) shows that for 0.01% of the time, the peaks now only exceed 7.4 dB instead of 9.2 dB. Essentially this means the signal’s maximum PAPR had been reduced by 1.8 dB for 99.99% of the time. Using the -3%/dB slope derived from Figure 6, this reduction in PAPR indicates an EVM degradation of approximately 5.4%.
Using a combination of peak power sensors and the CCDF lets you obtain rapid, near real-time results while adjusting IBO or other amplifier parameters. This allows you to find the optimum point on the amplifier’s linearity curve to balance IBO and PAE. In a production test, you need only monitor changes in PAPR to ensure you’re meeting EVM targets.
By leveraging a relationship between EVM and PAPR, you can measure PAPR reduction, which indicates EVM degradation, instead of expensive signal analyzers. Once you find the minimum level of PAPR, you can employ peak-power sensors to characterize PAPR and CCDF as a simple, fast, and cost-effective way to verify that you’ve attained the desired PAPR and hence EVM.
Tell Us What You Think!